Adapting the Neural Network Approach to PGA Prediction: An Example Based on the KiK-net Data

Derras, Boumédiène; Bard, Pierre‐Yves; Cotton, Fabrice; Bekkouche, Abdelmalek

doi:10.1785/0120110088

Cited by 96 publications

(49 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, in many fields of science, data-driven approaches have shown their suitability for generating models with high predictive powers (e.g., Breiman, 2001). In the context of GMPE calibration, examples of data-driven approaches include the application of artificial neural networks (e.g., Derras et al, 2012Derras et al, , 2014. To exemplify their application to the case study at hand, Figure 2 shows a simple neural network that includes one hidden layer with three neurons.…”

Section: Discussionmentioning

confidence: 99%

The Predictive Power of Ground‐Motion Prediction Equations

Bindi

2017

Bulletin of the Seismological Society of America

View full text Add to dashboard Cite

Although model calibration should be always performed alongside model validation, this is rarely the case when deriving new ground-motion prediction equations (GMPEs). Explanatory modeling (Shmueli, 2010) is often preferred to data-driven predictive approaches, and the analysis of the residual distribution is generally used to support the model qualification. Following previous studies (Kuehn et al., 2009;Scherbaum et al., 2009), this work aims to again stress the importance of validation for assessing the predictive power of GMPEs and for avoiding, or at least limiting, data overfitting. Considering the strong-motion data and models recently analyzed by Roselli et al. (2016), I exemplify the application of standard validation approaches based on predictive metrics (Akaike and Bayesian information criteria), resample techniques (cross validation and bootstrap), and validation against new data. The results confirm that GMPEs overfitting the calibration data have a limited predictive power, whereas, regarding validation against new data, the selection of the validation data set should take into account possible regional effects in the ground motion.

show abstract

Section: Discussionmentioning

confidence: 99%

The Predictive Power of Ground‐Motion Prediction Equations

Bindi

2017

Bulletin of the Seismological Society of America

View full text Add to dashboard Cite

show abstract

“…In our ANN models, we used this type of approach in order to facilitate the comparability with classical GMPEs. The ANN has the advantage that no prior functional form is needed (Derras et al 2012): the actual dependence is established directly from the data and can therefore be used as a guide for a better understanding of the factors which control ground motions. This resulted in a two-phase building process.…”

Section: Methodsmentioning

confidence: 99%

“…To avoid "overfitting" problems, we chose an adequate regularisation method involving the modification of the conventional mean sum of the squares of the network errors by the addition of a term equal to the sum of the squares of the network synaptic weights and bias (Derras et al 2012(Derras et al , 2014. Moreover, the optimal activation functions were found to be a "tangent sigmoïd" for the hidden layer and "linear" for the output layer.…”

Section: Fixed Modelsmentioning

confidence: 99%

“…Campbell (1989) found that adding a parameter for depth to basement rock improved the predictive ability of empirical ground-motion models. On their side, Cadet et al (2011) and Derras et al (2012) used another SCP: the fundamental resonance frequency, f 0 , as determined by the horizontal-to-vertical (H/V) spectral ratio technique (Mucciarelli 1998;Haghshenas et al 2008;Bard et al 2010). As the f 0 (H/V) SCP is able to identify low-frequency amplification on thick sites, its relevance may be compared with the performance of another SCP often proposed to properly account for the sediment thickness, H 800 (depth beyond which the shear-wave velocity exceeds 800 m/s).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

V S30, slope, H 800 and f 0: performance of various site-condition proxies in reducing ground-motion aleatory variability and predicting nonlinear site response

Derras

Bard

Cotton

2017

Earth Planets Space

View full text Add to dashboard Cite

The aim of this paper is to investigate the ability of various site-condition proxies (SCPs) to reduce ground-motion aleatory variability and evaluate how SCPs capture nonlinearity site effects. The SCPs used here are time-averaged shear-wave velocity in the top 30 m (V S30 ), the topographical slope (slope), the fundamental resonance frequency (f 0 ) and the depth beyond which V s exceeds 800 m/s (H 800 ). We considered first the performance of each SCP taken alone and then the combined performance of the 6 SCP pairs [. This analysis is performed using a neural network approach including a random effect applied on a KiK-net subset for derivation of ground-motion prediction equations setting the relationship between various ground-motion parameters such as peak ground acceleration, peak ground velocity and pseudo-spectral acceleration PSA (T), and M w , R JB , focal depth and SCPs. While the choice of SCP is found to have almost no impact on the median groundmotion prediction, it does impact the level of aleatory uncertainty. V S30 is found to perform the best of single proxies at short periods (T < 0.6 s), while f 0 and H 800 perform better at longer periods; considering SCP pairs leads to significant improvements, with particular emphasis on [V S30 -H 800 ] and [f 0 -slope] pairs. The results also indicate significant nonlinearity on the site terms for soft sites and that the most relevant loading parameter for characterising nonlinear site response is the "stiff" spectral ordinate at the considered period.

show abstract

“…Conversely, empirical estimation of NL site effects is more limited, especially in moderate seismicity areas where the onsite instrumental approach can only be a long (to very long)-term investment. Aside from a generic approach based on existing recordings (Derras et al, 2012;Sandıkkaya et al, 2013), the only presently possible way for site-specific estimates is thus numerical simulation. Obviously, such analysis must be preceded by a precise geotechnical and geophysical characterization of the underground structure, and the choice of a suitable NL constitutive model.…”

Section: Introductionmentioning

confidence: 99%

International Benchmark on Numerical Simulations for 1D, Nonlinear Site Response (PRENOLIN): Verification Phase Based on Canonical Cases

Régnier¹,

Bonilla²,

Bard³

et al. 2016

Bulletin of the Seismological Society of America

View full text Add to dashboard Cite

PREdiction of NOn-LINear soil behavior (PRENOLIN) is an international benchmark aiming to test multiple numerical simulation codes that are capable of predicting nonlinear seismic site response with various constitutive models. One of the objectives of this project is the assessment of the uncertainties associated with nonlinear simulation of 1D site effects. A first verification phase (i.e., comparison between numerical codes on simple idealistic cases) will be followed by a validation phase, comparing the predictions of such numerical estimations with actual strongmotion recordings obtained at well-known sites. The benchmark presently involves 21 teams and 23 different computational codes.We present here the main results of the verification phase dealing with simple cases. Three different idealized soil profiles were tested over a wide range of shear strains with different input motions and different boundary conditions at the sediment/bedrock interface. A first iteration focusing on the elastic and viscoelastic cases was proved to be useful to ensure a common understanding and to identify numerical issues before pursuing the nonlinear modeling. Besides minor mistakes in the implementation of input parameters and output units, the initial discrepancies between the numerical results can be attributed to (1) different understanding of the expression "input motion" in different communities, and (2) different implementations of material damping and possible numerical energy dissipation. The second round of computations thus allowed a convergence of all teams to the Haskell-Thomson analytical solution in elastic and viscoelastic cases. For nonlinear computations, we investigate the epistemic uncertainties related only to wave propagation modeling using different nonlinear constitutive models. Such epistemic uncertainties are shown to increase with the strain level and to reach values around 0.2 (log 10 scale) for a peak ground acceleration of 5 m=s 2 at the base of the soil column, which may be reduced by almost 50% when the various constitutive models used the same shear strength and damping implementation.

show abstract

Adapting the Neural Network Approach to PGA Prediction: An Example Based on the KiK-net Data

Cited by 96 publications

References 34 publications

The Predictive Power of Ground‐Motion Prediction Equations

The Predictive Power of Ground‐Motion Prediction Equations

V S30, slope, H 800 and f 0: performance of various site-condition proxies in reducing ground-motion aleatory variability and predicting nonlinear site response

International Benchmark on Numerical Simulations for 1D, Nonlinear Site Response (PRENOLIN): Verification Phase Based on Canonical Cases

Contact Info

Product

Resources

About